Acute kidney injury (AKI) has a negative impact on long-term renal function and prognosis. However, the association between acute renal dysfunction and long-term effects on bone disorders has not yet been characterized. Using a population-based cohort study, we aimed to evaluate associations between AKI and long-term effects on bone fractures. We identified relevant data of all hospitalized patients aged >18 years with histories of dialysis-requiring AKI, with subsequent recovery and discharge, from the claim records of the Taiwan National Health Insurance database between 2000 and 2008. We determined long-term de novo bone fracture and all-cause mortality after patients' index-hospitalization discharge using propensity score–adjusted Cox proportional hazard model. Varying-time models were used to adjust for long-term effects of end-stage renal disease (ESRD) on main outcomes. Among 448 AKI patients who had dialysis and survived 90 days after index-hospitalization discharge without reentering dialysis, 273 were male (60.9%) with a mean age of 61.4 ± 16.6 years. Controls included 1792 hospitalized patients without AKI, dialysis, or bone fracture history. In the AKI recovery group, bone fracture incidence was 320 per 10,000 person-years and hazard ratio (HR) of long-term bone fracture was 1.25 (p = 0.049) compared with the control group, independent of subsequent ESRD status (HR = 1.55; p = 0.01). Both AKI recovery status (HR = 2.31; p < 0.001) and time varying factor of bone fracture (HR = 1.43; p < 0.001) were independent predictors of mortality compared with controls. In conclusion, AKI requiring temporary dialysis independently increases long-term risk of bone fracture, regardless of subsequent progression to ESRD. Long-term bone fractures may negatively impact patient mortality. © 2014 American Society for Bone and Mineral Research.
The incidence of acute kidney injury (AKI) is increasing among hospitalized patients. Treating the condition consumes significant amounts of resources even after hospital discharge. Although AKI was once considered a reversible condition, mounting evidence has indicated that AKI may have a negative impact upon subsequent renal function and long-term prognosis, even if kidney function appears to be “recovered.”
Since the publication of the “Kidney Disease: Improving Global Outcomes (K-DIGO) Chronic Kidney Disease-Mineral Bone Disease (CKD-MBD) Guidelines,” there is a renewed interest in the influence of renal insufficiency on bone disorders. The inherent pathophysiologic connections between kidney and bone disease are diverse, including divalent ion imbalance, hormonal dysregulation, and the putative deleterious impact exerted by uremic toxins. To the best of our knowledge, the relationship between acute renal dysfunction and its long-term effect on bone disorders has not yet been characterized.
Therefore, we conducted a population-based study to investigate the association between AKI and its long-term extra-renal outcomes with a focus on bone fracture. We hypothesized that patients who survived and recovered from an episode of dialysis-requiring AKI during index admission would have higher long-term risk of bone fracture than other hospitalized patients without AKI.
Materials and Methods
The Taiwan National Health Insurance (NHI) database contains information from a nationwide insurance program covering outpatient visits, hospital admissions, prescriptions, interventional procedures, and disease profiles for more than 99% of the population in Taiwan (23.12 million in 2009).[5-7] It is one of the largest and most comprehensive databases in the world and has been used extensively in various studies investigating prescription use, diagnoses, and hospitalizations.
In cooperation with the Bureau of NHI, the National Health Research Institutes (NHRI) of Taiwan used a systematic, random sampling method to build a representative database of 1,000,000 NHI enrollees for the purpose of research.[5, 6] There are no statistically significant differences in terms of age, sex, or health-care costs between the sample group and all other enrollees captured in the NHI database, as reported by the NHRI. The sample data set used in our study spans from January 1, 2000, through December 31, 2008, and includes all claims data for 1,000,000 individuals. It is available to scientific researchers after encryption of all patients' personal information.
Informed consent was originally obtained by the NHRI, and because patients were anonymous in the present study, informed consent was not required. Also, because the identification numbers of all individuals in the NHRI databases were encrypted to protect the privacy of the individuals, this study was exempt from full ethical review by the institutional review board.
In this matched cohort study, all individuals who met the AKI case definition (according to the International Classification of Disease, 9th revision, Clinical Modification [ICD-9-CM] diagnosis and procedural codes, as provided in Supplemental Table S1) between January 1, 2000, and December 31, 2008, were extracted as the study population of interest. However, diagnosis of AKI is often underreported during admission if only AKI codes are identified using ICD. We, therefore, identified all AKI patients if their pre-admission procedural codes did not contain dialysis (procedure codes), but admission claimed data had included these codes. This method to identify AKI patients has been utilized before and has high accuracy because the Taiwan NHI database is claim based and procedural codes are always linked to reimbursements, which would not be missed by physicians. Individuals' comorbidities were identified and confirmed at the start of the illness claim codes within 1 year before the index admission. Cases were defined as patients who recovered from temporary dialysis within 90 days after index-hospitalization discharge. To minimize selection bias, propensity scores were calculated to adjust for covariates and baseline differences between the two study groups. The sampling ratio between patients with and without AKI was 1:4. Computer-generated random sampling was used to select patients for the control group. Controls were hospitalized patients without diagnosis of AKI. Patients with a prior history of renal transplantation, length of hospital stay exceeding 180 days during the index hospitalization, with evidence of prior dialysis treatment, or those with any records of bone fracture in the preceding 1 year before index admission were excluded.
Comorbidities were defined according to ICD-9-CM and procedure codes (including Taiwan Classification of Procedures) (Supplemental Table S1). Charlson comorbidity index scores were calculated based on preexisting conditions identified from each patient's medical records. Demographic and clinical characteristics of study subjects (including age, sex, admission year, and presence of comorbidities) during index hospital admission were examined. Data pertaining to the index hospitalization, including diagnosis codes, categories of major operations, resources used (including hemodialysis), mechanical ventilation (MV), intensive care unit (ICU) admission, and patient outcomes were retrieved. Preexisting comorbidities were defined as any treated diseases with at least one hospital admission or three outpatient visits in the preceding 1 year before index admission. The extraction of presence of covariates was available from both the index hospitalization and clinics visited at 1 year before hospitalization data and was well validated with good predicting power.[5, 12] Specifically, comorbid CKD is defined according to ICD-9-CM coding rather than to estimated glomerular filtration rate (eGFR). This method demonstrated fair validity in estimating CKD prevalence and has also been utilized in past reports.[5, 13] According to Taiwan NHI, patients with renal anemia and CKD, with serum creatinine levels higher than 6 mg/dL, are eligible for prescriptions of erythropoiesis stimulating agents (ESAs) without copayment. Consequently, we further selected patients with “advanced CKD,” defined as the presence of CKD coding and concomitant reimbursement coding of ESAs without copayment, in all study groups to differentiate the degree of renal dysfunction within groups. Because a predialysis multidisciplinary care program has been put forth by the Department of Health in Taiwan, with a very high coverage rate, the percentage of reimbursed use of ESAs in those eligible CKD patients could be very high, too.
Definition of outcomes
We followed up each enrollee for any bone fracture events until December 31, 2009. Patients were prospectively followed up for at least 1 year after their discharge. The primary endpoints were defined as the first reported bone fracture event, whereas secondary endpoints were defined as ESRD or all-cause mortality, as defined by the ICD-9-CM codes within the hospitalization or outpatient data.[14, 15] The duration between the day of discharge from index admission to the first reported fracture event or ESRD was defined as “bone event time” and “ESRD event time,” respectively. The codes of bone fractures were validated and reported. In Taiwan, patients who continue dialysis for more than 90 days all apply to the NHI for catastrophic illness registration cards, which ensures the accuracy of our definition of dialysis continuation. As in our previous study, we used a selection period of 90 days to reduce the possibility of survivor-treatment bias.
Continuous variables were reported as mean ± standard deviations; discrete variables were presented as counts or percentages. All data were analyzed using R software, version 2.14.1 (Free Software Foundation, Inc., Boston, MA, USA). Cox proportional hazard regression model and propensity scoring were used to conduct separate analyses within each stratum and evaluate the risk of events after adjustments for all variables. Two-sided p values < 0.05 were considered statistically significant.
Propensity score matching (PSM) was used to balance the covariates in the case-control study to reduce the bias and enable estimation of an average treatment effect from observational data. PSM, at the time of its introduction, allowed researchers to balance treatment and control groups and simulated a randomized control by matching a pair of patients with similar propensity score to eliminate confounding factors. We used logistic regression to estimate propensity scores to match characteristics from each group of subjects, including age, sex, pre- and index hospitalization comorbidities, and operative categories as listed in Table 1.
|Recovery Group (n = 448)||Non-AKI Group (n = 1792)||p Value|
|Male||273 (60.9%)||1113 (62.1%)||0.664|
|Age (years)||61.4 ± 16.6||63.5 ± 15.4||0.001|
|Charlson score||1.9 ± 1.9||1.9 ± 1.9||0.713|
|Myocardial infarction||8 (1.8%)||61 (3.4%)||0.091|
|Congestive heart failure||34 (7.6%)||142 (7.9%)||0.922|
|Peripheral vascular disease||6 (1.3%)||16 (0.9%)||0.420|
|Cerebrovascular disease||39 (8.7%)||139 (7.8%)||0.495|
|Dementia||13 (2.9%)||56 (3.1%)||0.880|
|COPD||44 (9.8%)||179 (10%)||0.999|
|Rheumatologic disease||6 (1.3%)||28 (1.6%)||0.832|
|Peptic ulcer||58 (12.9%)||260 (14.5%)||0.449|
|Hemiplegia||6 (1.3%)||23 (1.3%)||0.999|
|Solid tumor||28 (6.2%)||125 (7%)||0.675|
|Metastatic tumors||13 (2.9%)||43 (2.4%)||0.503|
|Diabetes mellitus||149 (33.3%)||577 (32.2%)||0.693|
|Moderate or severe liver disease||26 (5.8%)||108 (6%)||0.912|
|Chronic kidney disease||128 (28.6%)||513 (28.6%)||0.999|
|Advanced CKD||24 (5.4%)||81 (4.5%)||0.316|
|Index hospital comorbidity|
|Cardiothoracic||23 (5.1%)||61 (3.4%)||0.095|
|Respiratory||45 (10%)||159 (8.9%)||0.463|
|Hepatic||11 (2.5%)||33 (1.8%)||0.445|
|Neurologic||4 (0.9%)||11 (0.6%)||0.518|
|Hematologic||4 (0.9%)||6 (0.3%)||0.120|
|Metabolic||3 (0.7%)||10 (0.6%)||0.731|
|Cardiothoracic||15 (3.3%)||70 (3.9%)||0.679|
|Upper GI||5 (1.1%)||24 (1.3%)||0.819|
|Lower GI||8 (1.8%)||18 (1%)||0.213|
|Hepatobiliary||6 (1.3%)||30 (1.7%)||0.833|
|Mechanical ventilation||130 (29%)||194 (10.8%)||<0.001|
|ICU admission during index hospitalization||225 (50.2%)||276 (15.4%)||<0.001|
Because of the strong correlation between ESRD and bone fragility,[20, 21] we used a Cox proportional hazards model with time-varying covariates to evaluate the impact of ESRD on the risk of bone fracture, assuming that changes in ESRD status could appear at any time between the index-hospitalization discharge and the bone fracture event. In addition, to address the complex effects of age, prior CKD, AKI, subsequent ESRD, and their impact on the long-term bone fracture risk, we further added interaction variables including age-by-AKI, age-by-ESRD, AKI-by-subsequent-ESRD, and AKI-by-prior-CKD into our analysis. Furthermore, a conditional effect plot for outcome prediction was drawn based on the fitted final logistic regression model. The model estimated probability of having an unfavorable outcome against a chosen continuous covariate, with the values of the other discrete and continuous covariates held constant.
A total of 11,766 patients who were diagnosed with dialysis-requiring AKI between January 1, 2000, and December 31, 2008, during their index hospitalization were included. We excluded 769 patients who had ICD-9-CM codes of bone fracture during the preceding year before hospital discharge. Another 6227 patients were excluded from participation because of having dialysis or AKI within the past year, history of renal transplantation, or prolonged hospital stay (>180 days) during index admissions. Patients who had expired during the index hospitalization (n = 684) or within 3 months after hospital discharge (n = 1298) were also excluded. Finally, 2788 patients with dialysis-requiring AKI during the index hospitalization were identified. A total of 661 patients remained dialysis-independent for at least 3 months after discharge. However, 213 patients were not propensity score-matched and were excluded. Thus, 448 patients were included in the study group of interest (AKI recovery group). Of these, 273 were male (60.9%) with a mean age of 61.4 ± 16.6 years. In addition, 1792 propensity score-matched patients without acute kidney injury or dialysis in the past year and during hospitalization were included as the non-AKI control group, with a 1:4 ratio.
Patients from the AKI recovery group were younger than those without AKI during hospital admissions (61.39 ± 16.55 versus 63.46 ± 15.38; p = 0.001). No significant differences were found between the two groups in terms of their comorbidities and causes for admission. The proportion of “advanced CKD” in each group also did not differ (p = 0.316). The AKI recovery group was more likely to receive mechanical ventilation (p < 0.001) and ICU transfer (p < 0.001) than the non-AKI group. During a mean follow-up period of 3.9 ± 2.6 years, a significantly greater proportion of patients from the AKI recovery group developed ESRD compared with patients in the non-AKI group (19% versus 1.2%; p < 0.001).
Characteristics of bone fracture events
The sites of fractures are presented in Table 2. Most fractures occurred in either vertebrae (3.1%) or femoral neck (2.9%), whereas fractures in the smaller appendicular bones (carpal/tarsal bones or phalanges) were less frequent.
|Anatomical sites||Recovery Group (n = 448)||Non-AKI Group (n = 1792)|
|Vertebral||14 (3.1%)||69 (3.9%)|
|Pelvic||2 (0.4%)||11 (0.6%)|
|Clavicular or scapular||7 (1.6%)||9 (0.5%)|
|Humeral||8 (1.8%)||29 (1.6%)|
|Radial or ulnar||4 (0.9%)||35 (2%)|
|Carpal, metacarpal, or phalanges||2 (0.4%)||22 (1.2%)|
|Femoral (neck)||14 (3.1%)||64 (3.6%)|
|Femoral (other sites)||7 (1.6%)||27 (1.5%)|
|Patellar||1 (0.2%)||6 (0.3%)|
|Tibial or fibular||7 (1.6%)||20 (1.1%)|
|Ankle||6 (1.3%)||9 (0.5%)|
|Tarsal, metatarsal, or phalanges||2 (0.4%)||20 (1.1%)|
Long-term risk of bone fractures
Additional adjustment in the multivariate model was made for variables, including age, sex, and comorbidities, as listed in Table 1. The propensity scores for the AKI recovery group were subsequently calculated to minimize residual confounding effects beyond the matching process.
The incidence of bone fracture was 320 per 10,000 person-years in the AKI recovery group and 93 per 10,000 person-years in the non-AKI recovery group. The hazard ratio (HR) for developing bone fracture in the AKI recovery group, relative to non-AKI patients, was 6.59 (95% confidence interval [CI] 2.45–17.73; p ≤ 0.001), corresponding with an absolute risk reduction of 227 per 10,000 person-years (Table 3).
|HR||95% CI Lower Limit||95% CI Upper Limit||p Value|
|Peripheral vascular disease||2.49||1.41||4.39||<0.001|
|Upper GI surgeries||2.84||1.50||5.36||<0.001|
|Recovery group versus non-AKI subgroup||6.59||2.45||17.73||<0.001|
|AKI × age||0.98||0.96||0.99||<0.001|
|AKI × ESRD||1.49||1.05||2.11||0.02|
Other risk factors associated with the development of fractures included age (HR = 1.04 per year; p < 0.001), female gender (HR = 1.27; p = 0.02), presence of peripheral vascular disease (HR = 2.49; p < 0.001), chronic obstructive pulmonary disease (HR = 1.58; p < 0.001), diabetes mellitus (HR = 1.43; p < 0.001), rheumatologic diseases (HR = 2.02; p = 0.03), and neurologic diseases (HR = 3.96; p < 0.001) (Table 3). In addition, borderline higher risk was indicated among patients who underwent cardiothoracic surgery (HR = 1.79; p = 0.05), while significantly higher among those undergoing upper gastrointestinal (GI) surgery (HR = 2.84; p < 0.001) during index admission. Finally, interaction variables including AKI-by-age (HR = 0.98; p < 0.001) and AKI-by-subsequent ESRD (HR = 1.49; p = 0.02) were shown to be significantly associated with long-term fracture risk. The constructed model demonstrated good validity (C-index = 0.68).
There were 213 (32.3%) participants with acute kidney injury and dialysis who could not be matched to a suitable control by propensity score matching. The respective incidence rate of bone fracture in this group was 422.3 per 10,000 person-years, which was higher than the rate for the 448 patients who were successfully matched.
Sensitivity analyses were done, focusing on the weight-bearing bones as the clinical outcomes and also the distribution of renal dysfunction severity in each group. After excluding fractures of clavicles, scapular, fingers, and toes, the recovery group still had higher long-term risk for developing bone fracture than non-AKI patients (HR = 6.02, 95% CI 1.43–17.22; p = 0.02). In addition, the study results did not change significantly after excluding patients with predefined advanced CKD.
Conditional effect plots stratified by gender and study groups
To further explore the association of advanced age with the incidence of bone fracture, we constructed a statistical correlation curve based on the aforementioned results. Fig. 2 illustrates a conditional effect plot of the estimated risk for long-term fractures, stratified by AKI status and gender, against patients' ages, whereas the other factors were held constant. Results suggest that, in the long term, female patients with AKI requiring temporary dialysis had a significantly higher incidence of bone fracture compared with those without AKI during their index admission. In addition, the differences in fracture risk between the AKI and non-AKI groups significantly increased as patients' ages rise (with transition at 62 years) (Fig. 2).
Bone fracture related to long-term mortality
Time-varying Cox regression model found that long-term mortality was associated with both AKI recovery status (HR = 2.31, 95% CI 2.07–2.59; p < 0.001) and time-varying bone fracture (HR = 1.43, 95% CI 1.19–1.71; p < 0.001). This model showed good validity, with a C-index of 0.73.
In the present study, using a nationwide population-based database, we found that recovery from dialysis-requiring AKI carries a significantly higher risk of developing subsequent bone fractures after hospital discharge. Moreover, the bone fractures per se have a significant impact on patients' long-term mortality. These findings are important for clinicians who provide long-term care to patients who have ever had episodes of dialysis-requiring AKI, especially if the patients present with bone fractures during follow-up. To the best of our knowledge, this relationship has not been explored previously. We believe that these results may shed light on the impact of AKI on the development of bone diseases.
In this population-based study, long-term mortality is positively associated with episodes of bone fracture, independent of previously temporary dialysis, in accordance with several previous reports concerning hip and vertebral fractures. Therefore, although the renal insults play a role in affecting long-term patient mortality, the results from the present study additionally offer an insight into the impact of bone disease. Furthermore, we propose that preventive strategies on bone fractures in AKI patients could have a great impact on patient mortality.
In general, the incidence of bone fractures in Taiwan was reported to be 60 per 10,000 person-years for all ambulatory patients, which is comparable to the incidence in other developed countries.[25, 26] The etiology of bone fractures includes musculoskeletal injury or trauma and, more importantly, osteoporosis associated with increasing population age. In light of these findings, it is important to clearly define the potential factors that might contribute to higher risk of fracture events. In this study, we found that the incidence of bone fractures was 320 per 10,000 person-years among patients in the AKI recovery group. This number is higher than that in the general population. A potential explanation of these differences may be related to the higher biologic ages in both groups and the higher number of complicated cases recruited into the study population (more ICU admissions and ventilation rates) (Table 1).
The development of subsequent ESRD after a recovered AKI episode raises questions about the contribution of AKI on excess fracture risk. After adjustment of ESRD as a covariate in our study model, results still demonstrated that AKI is independently associated with increased risk of bone fractures, irrespective of disease progression to ESRD. Furthermore, baseline CKD could also be contributory to the enhanced risk of subsequent fractures and precipitate AKI.[28, 29] We also adjusted baseline CKD in our survival modeling, and AKI recovery status had outperformed baseline CKD in elevating risk of late fractures (Table 3). Finally, the fact that a baseline higher biologic age in non-AKI group than AKI recovery group (63.5 versus 61.4 years; p = 0.001; Table 1) did not lead into higher risk of fractures during follow-up further enhances the biologic plausibility of our findings.
Our results also showed that progression to ESRD is associated with higher risk of bone fractures, which is further compounded by the effect of aging. This finding is consistent with other studies, which showed that AKI survivors have increased long-term mortality and risk of developing CKD/ESRD.[2, 3] The incidence of hip fractures in ESRD is three- to fourfold higher than in the general population. Contrary to CKD/ESRD, literature on the effects of acute kidney dysfunction on bone structure and fracture is extremely sparse. Because AKI is gradually being recognized as a contributing factor to late-stage chronic kidney disease (CKD), it is possible that early kidney changes associated with CKD may have already taken place during acute events, paving the way toward progressive deterioration. In our study, the AKI recovery group had a high risk for long-term bone fracture in addition to subsequent ESRD, representing a new mechanism that was previously unidentified. Furthermore, our sensitivity analysis showed that AKI patients had a high incidence of developing fractures of weight-bearing bones. This could potentially be explained by the fact that complications of CKD, including deranged mineral hormone axis and ion concentration, are also found during AKI. A recent report by Leaf and colleagues identified a prominent elevation of FGF-23 among patients with AKI, and higher levels correlated with adverse outcomes. The same study also found that the vitamin D metabolite, 25-hydroxy- and 1,25-hydroxy-vitamin D, was lower in AKI patients. Another study by Zhang and colleagues also discovered that severe hyperparathyroidism occurs in patients who suffer from AKI, with a striking elevation of FGF-23 and hyperphosphatemia. Because pathological changes of renal osteodystrophy reportedly start earlier than expected CKD stages and dysregulated mineral hormones also occur in AKI, a renowned predecessor of CKD, it is plausible that earlier changes in vitamin D metabolites/FGF-23 levels during AKI could translate into subsequent bone structural abnormalities. Several recent reports have also suggested that FGF-23 level could be a predictor of long-term vertebral fracture events.[35, 36] However, the role of FGF-23 in skeletal mineralization status is still unclarified and remains to be determined for AKI patients. In addition to the hormonal influence on bone structure, uremic toxins might also play a role in promoting bone abnormalities. The accumulation of uremic toxins in AKI might carry the potential of disturbing bone remodeling and causing structural defects, resulting in bone fracture even after AKI recovery. Nonetheless, the exact pathogenic mechanisms involving AKI and subsequent fractures are still unknown and warrant further investigation.
Another interesting finding in our study is that after adjusting for relevant factors, the risk carried by temporary dialysis seems to increase further in older age (Fig. 2). Hip fractures were the most common anatomical sites of fractures, and the increase of risk distributed equally in both study groups. Osteoporosis is a well-established risk factor for bone fractures among older adult female patients. As demonstrated in this study, female patients with AKI who had temporary dialysis had greater incidence of bone fracture than did males with AKI and temporary dialysis. This finding is comparable to the results of another study that reported apparent gender differences in the development of mineral bone disease (MBD), which were accentuated by age. Although age per se already increases the risk of fractures, the clinical implication of AKI-associated long-term fractures is still vital: In an ever-growing aging population (as in most countries currently), the impact of AKI on the subsequent risk of fractures could be more devastating in the geriatric population than among young adults, even if the AKI episodes have seemingly improved. In light of this information, a comprehensive rehabilitation plan is required during admission or even after discharge for older adult patients who have had AKI to reduce the risk of ensuing bone fractures.
Other risk factors identified in the current study are comparable to those in other published studies, including age, diabetes mellitus, and rheumatologic diseases. COPD presumably increases the risk of fractures resulting from the frequent use of inhaled or systemic corticosteroids. The presence of respiratory failure, neurologic diseases, and prior cardiothoracic surgery are all associated with increased risk of developing AKI and subsequent fractures, which are potentially complicated by prolonged physical inactivity owing to patients' bed-bound status.[42-44] Peripheral vascular disease has also been reported to increase risk of hip and nonvertebral fractures, but our findings showed that the risk is higher than the previous study (Table 2).
The ethnic differences in fracture risks determination also deserve attention during result interpretations. Asians have, on average, lower bone mineral density (BMD) compared with whites, but the incidences of hip or wrist fractures are lower than whites paradoxically.[46, 47] Patients of African descent have the highest BMD among all ethnic groups, but white patients have the highest risk of fractures. The explanation for the racial differences in fracture risks might be partially attributed to BMD variation, but factors such as body weight, body height, or specific bone morphometric features could also be responsible. Consequently, we should bear in mind the potential inherent ethnic differences before generalizing our findings to other populations.
Our study had several limitations. First, there was no corresponding BMD data for comparison. Because few patients with AKI underwent a dual-energy X-ray absorptiometry (DXA) scan, etiology of fractures could not be investigated accordingly. Second, other residual factors may have been present that could influence our analysis and results, such as body mass index (BMI). BMI could be an important determinant of BMD, and being obese or underweight reportedly increases fracture risk, but BMI data were unavailable in our study. Also, the baseline GFR was not available from the ICD codes, and misclassification was possible. However, such misclassification was likely to be random and thus would tend to underestimate rather than overestimate the association. Similarly, RIFLE classification of AKI could not be applied in our study because serum creatinine levels were unavailable.
The coding accuracy of ICD-9-CM for fractures has been debated. In Sweden, an external review and validation by Swedish national inpatient registry for fracture events by national quality registries showed that 94% to 96% of cases of knee and hip fractures were correctly identified. Another Canadian validation study for fracture codings also showed that for wrist, humerus, and vertebral fractures, trend estimates of ICD-identified cases and clinical-identified ones were similar, whereas hip fractures displayed gender-specific disparity in ICD-identified cases. However, other studies suggested that ICD codes could be inaccurate in identifying cases of fracture. A plausible explanation for this controversy lies in the variations in each study regarding fracture case definitions, ICD code compositions, anatomical sites of interest, and even the specific subareas of each fracture site. Nonetheless, the reimbursement practice in Taiwan NHI required registration of both diagnostic and treatment codes, and thus the coding is likely more accurate than other studies. However, our sensitivity analysis proved that the impact of AKI on fracture persisted even if we considered only weight-bearing bone fractures (often more severe episodes), and this finding lent further support to the credibility of our results, although in our post hoc analysis, the coding of fractures at patella and other femoral sites from index hospitalization showed mildly higher performance than outpatients (Supplemental material).
Past reports also suggested that clinical vertebral fractures are frequently underreported in practice, especially among younger patients with fewer comorbidities. Asymptomatic vertebral fractures are not infrequent and could not be captured by ICD codes. This might also affect the results of our study. However, in our original cohort, patients with non-AKI were significantly younger than those with AKI recovery (Table 1). From this point of view, the finding of significantly higher risk of fractures in the AKI recovery group, which were more likely to have underreported vertebral fractures (Table 3), could lend support to the credibility of our results. The long-term effects of AKI on the development of bone fractures still warrants further research. Monitoring of MBD among patients with AKI may also help to validate our study findings.
In this study, we addressed an unexplored and underestimated issue of bone fractures after AKI. We found that even after patients recover from AKI requiring dialysis, they still have an increased risk of developing fractures, which has a further corresponding impact on long-term mortality. Therefore, our findings are among the first to demonstrate an important link between temporary dialysis after AKI and bone fractures. This is an important finding and may have implications in clinical management and care planning in patients with dialysis-requiring AKI.
All authors state that they have no conflicts of interest.
The authors thank the International Harvard Statistical Consulting Company, Taipei, and the staff of the Second Core Lab of the Department of Medical Research at National Taiwan University Hospital for technical assistance.
This study was supported by the Taiwan National Science Council (grants NSC 101-2314-B-002-132-MY3, NSC100-2314-B-002-119, NSC 101-2314-B-002-085-MY3) and by NTUH 100-N1776, 101-M1953, 102-S2097.
Authors' roles: Study design: YCH, CTC, WJW, and VCW. Study conduct: YCH, CTC, WJW, and VCW. Data collection: VCW. Data interpretation: YCH, CTC, WJW, VCW, WJK, and KDW. Drafting: YCH, CTC, WJW, VCW, WJK, and KDW. Revising manuscript content: YCH, CTC, WJW, VCW, WJK, and KDW. Approving final version of manuscript: YCH, CTC, WJW, VCW, WJK, and KDW. VCW was responsible for the integrity of the data analysis.